COS 60-10
Multiple sources of uncertainty affect commonly used metrics of conservation risk under climate change

Wednesday, August 13, 2014: 11:10 AM
Regency Blrm B, Hyatt Regency Hotel
Amber N. Wright, Biology, University of Hawaii at Manoa, Honolulu, HI
Robert J. Hijmans, Environmental Science and Policy, University of California, Davis, Davis, CA
Mark W. Schwartz, Department of Environmental Science and Policy, University of California, Davis, Davis, CA
H. Bradley Shaffer, Ecology and Evolutionary Biology, University of California - Los Angeles, Los Angeles, CA

Ecological niche models are one of the main tools for conservation planning under future climate change. However, these models can produce complicated outputs that are difficult to incorporate into the planning process. A commonly employed approach is to collapse ecological niche model results into metrics that rank species by degree of predicted future habitat loss. While many studies have considered how different modeling decisions contribute to uncertainty in model outputs, here we consider specifically effects on commonly used ranking metrics. We built ecological niche models for 153 species of reptiles and amphibians in California, USA. We used AICc to select climate variables and tune Maxent’s built-in regularization parameter to build reduced complexity models for comparison with default models. We predicted the distribution of climatically suitable habitat under future (2041-2060) climate conditions according to 11 global climate models and 4 representative concentration pathways. We calculated two ranking metrics, a “no dispersal” metric and a “some dispersal” metric using three different threshold values to determine suitable vs. unsuitable habitat. We used general linear mixed models to determine the effects of modelling decisions on rankings.


We found that while individual modeling decisions had relatively small effects on species ranks alone, in combination these decisions can lead to very different conservation assessments. Most species received 4 or more different ranks across all conditions considered. Under the most optimistic combination of modeling decisions, 89% of species received the same rank and were predicted to experience mild changes in climatically suitable area of +/- 20%. At the other extreme, the distribution of ranks across species was uniform, indicating large increases and decreases in future habitat. Reducing model complexity decreased predicted conservation risk assessed at currently occupied localities but not when considering shifts in suitable area within species ranges. Overall, even though ranks are relatively coarse in scale, they are still affected by multiple sources of uncertainty in the modeling process. We recommend that a wide range of modeling decisions be explored and that variation in ranks across runs be reported as a first step in acknowledging uncertainty in rank metrics used for assessing conservation risk under climate change.